A Method to Recognize Significant Informative Speech Signal Parameters Relevant to Human Emotional States

At present, various physiological signals that capture electrical activity from the human body (electrocardiography, electroencephalography, oculography, etc.) are used to detect and classify emotional states. Since the speech apparatus is sensitive to the nervous system disorders, human emotional states and emotions are encoded into relevant speech parameters contingent upon the psychological situation. This paper aims to assess the reliability of recognizing significant informative emotional speech parameters, and the degree of relevance thereof to the emotions experienced by a person. The scientific novelty of the work is to determine the threshold values for critical human emotional states under extreme conditions. The article proposes a method to recognize significant informative speech signal parameters using energy, prosodic, format, spectral, timing, cepstral, and nonlinear Teager energy operator-based analyzes. The method provides the possibility of expanding the information space for speech signal parameters through the use of a mathematical representation model based on the empirical mode decomposition modifications. The proposed method was investigated using a speech signal database of human negative (fear, disgust, anger, and sadness), neutral, and positive (surprise, joy) emotional states recorded in human real-life experiencing emotions. The research results have evidenced that the proposed method provides high-level relevance to human natural emotional states. The research findings have revealed the minimum and maximum recorded changes in the values of significant informative parameters for negative and positive states relative to a conditionally exemplary neutral state (a decrease by 18% and an increase by 40%, respectively).